Acceleration Method, Device, Electronic Device and Storage Medium for Ensemble Kalman Filter Algorithm Assimilation Calculation
By dividing the calculation area into multiple checkerboard sub-regions and performing assimilation calculations in parallel, the problem of inefficient calculation in the ensemble Kalman filtering algorithm is solved, and a more efficient calculation process is achieved, avoiding mutual interference between observation data and improving the efficiency of multi-core computing.
Patent Information
- Application Number
- CN202210764333.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-29
- Publication Date
- 2025-07-04
- Estimated Expiration
- 2042-06-29
AI Technical Summary
The existing ensemble Kalman filtering algorithm has problems such as low computational efficiency and difficulty in solving matrix in large-scale assimilation calculations, especially in parallel calculations, which are prone to interference with each other, resulting in a decrease in computational efficiency.
The calculation area is divided into multiple checkerboard sub-regions according to the influence radius of observation data, and the assimilation calculation of non-adjacent sub-regions is performed in parallel, and the adjacent sub-regions are updated synchronously until all sub-regions are completed.
More efficient parallel computing is realized, computing time is shortened, computing efficiency is improved, and observation data is avoided in the mutual interference of multi-core calculations, and there is no need to improve the ensemble Kalman filtering algorithm.
Smart Images
Figure CN115114583B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to data processing technologies, and more particularly, to an acceleration method, device, electronic device, and storage medium for assimilative calculation of an ensemble Kalman filter algorithm. Background Art
[0002] Data assimilation is a method for processing, adjusting, and integrating data from different sources to obtain an optimal state variable at the current time and space. According to the theory of data assimilation, it can be divided into two categories: one is based on statistical estimation theory, such as optimal interpolation, Kalman filtering, Kalman smoothing, etc.; the other is based on variational methods, such as 3DVAR, 4DVAR, etc. The ensemble Kalman filter (ENKF) algorithm is a typical filtering algorithm based on statistical estimation theory.
[0003] The Ensemble Kalman Filter (EnKF) is developed based on the stochastic dynamic prediction theory proposed by Epstein in 1969 (Epstein E S. Stochastic dynamic prediction[J]. Tellus, 1969, 21(6): 739-759). It was not until 1994 that the EnKF algorithm was first applied to data assimilation by Geir Evensen, and its application scenario was data assimilation in the ocean field (Evensen G. Sequential data assimilation with a nonlinear quasi-geostrophic model using Monte Carlo methods to forecast error statistics[J]. Journal of Geophysical Research: Oceans, 1994, 99(C5): 10143-10162). The EnKF designed by Evensen is a four-dimensional assimilation method that uses a Monte Carlo short-term ensemble prediction method to estimate the forecast error covariance. In 1998, Houtekamer and Mitchell further extended the EnKF method and introduced it into the atmospheric data assimilation system (Houtekamer P L, Mitchell H L. A sequential ensemble Kalman filter for atmospheric data assimilation[J]. Monthly Weather Review, 2001, 129(1): 123-137). They found that the use of multiple ensembles can reduce the inbreeding problem in the analysis process; the more accurate the estimation of the correlation of the number of ensemble members, the smaller the root mean square error of the analysis; it is difficult to accurately estimate the small correlation far from the observation, and there is false correlation noise; for a fixed number of ensemble members, there is an optimal truncation radius, and as the number of ensemble members increases, the optimal truncation radius increases.
[0004] When applying the ensemble Kalman filter method, as the number of ensemble members increases, the truncation radius increases, and the number of assimilation data increases, the matrix scale obtained by the ensemble Kalman algorithm becomes increasingly large, and the solution of the assimilation calculation becomes increasingly difficult. In addition, since the number of ensembles is much smaller than the dimension of the observations, the Kalman gain matrix is rank-deficient, and a rank-deficient matrix does not have an inverse matrix in mathematics, which also increases the difficulty of the solution. To overcome these difficulties, the commonly used method is the sequential assimilation method. Considering that when the observation data are uncorrelated with each other, the point-by-point sequential assimilation of the observation data is equivalent to the simultaneous assimilation. If a single-point (single station, single variable, single layer) sequential assimilation method is adopted, the matrix inversion becomes the reciprocal of a scalar. Although this sequential assimilation method overcomes the computational difficulties and the problem of the rank-deficiency of the Kalman gain matrix, due to the huge number of observation data and the dependence between the observation data, it is difficult to parallelize the calculation process.
[0005] To improve the computational efficiency, the parallel computing methods proposed by Keppenne et al. try to simultaneously input observation data that are far apart as much as possible, and utilize the characteristic that the observations far apart have no dependence to achieve parallelism and improve the computational efficiency (Keppenne C L, Rienecker M M. Assimilation of temperature into an isopycnal ocean general circulation model using a parallel ensemble Kalman filter[J]. Journal of Marine Systems, 2003, 40: 363-380). However, this will make the distribution highly random. Especially as the number of cores increases, when more observation data are simultaneously input, it is inevitable to include observation data that affect each other. At this time, the observation data that affect each other can only be processed serially, reducing the parallel efficiency. Summary of the Invention
[0006] The technical problem to be solved by this application is to provide an acceleration method, an acceleration device, an electronic device, and a computer-readable storage medium for the assimilation calculation of an ensemble Kalman filter algorithm that can effectively improve the computational efficiency in view of the above-mentioned defects of the prior art.
[0007] In a first aspect, this application proposes an acceleration method for the assimilation calculation of an ensemble Kalman filter algorithm to solve its technical problem. The method includes the following steps:
[0008] S11. Divide the calculation area into multiple sub-areas according to the influence radius of the observation data;
[0009] S12. Select a set of non - adjacent sub - regions from the multiple sub - regions, perform the assimilation calculation of the set of non - adjacent sub - regions in parallel, and synchronously update the adjacent sub - regions of each non - adjacent sub - region. Repeat the execution until the assimilation calculation of all the multiple sub - regions is completed.
[0010] In an embodiment of the first aspect of the present application, the length from the center of each sub - region in step S11 to its boundary is not less than the influence radius of the observation data.
[0011] In an embodiment of the first aspect of the present application, in step S11, the calculation region is divided into multiple square sub - regions, and the side length of each square sub - region is equal to the influence diameter of the observation data.
[0012] In an embodiment of the first aspect of the present application, step S12 includes:
[0013] Starting from the first sub - region of the coordinate system of the calculation region, select a set of non - adjacent sub - regions to perform the assimilation calculation in parallel;
[0014] Starting from the adjacent sub - regions of the first sub - region, select other sets of non - adjacent sub - regions to perform the assimilation calculation in parallel until the assimilation calculation of all the multiple sub - regions is completed.
[0015] In an embodiment of the first aspect of the present application, in step S12, the sequential assimilation calculation or parallel assimilation calculation is performed on the observation data inside each sub - region, and the adjacent sub - regions are synchronously updated.
[0016] In the second aspect, the present application proposes an acceleration device for the assimilation calculation of the ensemble Kalman filter algorithm to solve its technical problems, which is characterized by including:
[0017] A division module, configured to divide the calculation region into multiple sub - regions according to the influence radius of the observation data;
[0018] A parallel calculation module, configured to select a set of non - adjacent sub - regions from the multiple sub - regions, perform the assimilation calculation of the set of non - adjacent sub - regions in parallel, and synchronously update the adjacent sub - regions of each non - adjacent sub - region. Repeat the execution until the assimilation calculation of all the multiple sub - regions is completed.
[0019] In an embodiment of the second aspect of the present application, the length from the center of each sub - region to its boundary is not less than the influence radius of the observation data.
[0020] In an embodiment of the second aspect of the present application, the division module divides the calculation region into multiple square sub - regions, and the side length of each square sub - region is equal to the influence diameter of the observation data.
[0021] In an embodiment according to the second aspect of the present application, the parallel computing module is further configured to start from the first sub-region of the coordinate system of the computing region to select a set of non-adjacent sub-regions to perform assimilation calculations in parallel, and start from each adjacent sub-region of the first sub-region to select other sets of non-adjacent sub-regions to perform assimilation calculations in parallel until the assimilation calculations of all the multiple sub-regions are completed.
[0022] In an embodiment according to the second aspect of the present application, the parallel computing module performs sequential assimilation calculations or parallel assimilation calculations on the observation data within each sub-region, and synchronously updates adjacent sub-regions.
[0023] In a third aspect, to solve its technical problems, the present application provides an electronic device, including a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the acceleration method for the ensemble Kalman filter algorithm assimilation calculation as described above.
[0024] In a fourth aspect, to solve its technical problems, the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the acceleration method for the ensemble Kalman filter algorithm assimilation calculation as described above.
[0025] Implementing the acceleration method, acceleration device, electronic device, and computer-readable storage medium for the ensemble Kalman filter algorithm assimilation calculation of the present application has the following beneficial effects: Based on the characteristics of the observation data and taking the influence radius of the observation data as a benchmark, the present application divides the computing region into multiple checkerboard-shaped sub-regions, and then simultaneously performs assimilation calculations on a set of non-adjacent sub-regions, thereby achieving more effective parallel computing, shortening the computing time, and improving the computing efficiency. Moreover, the present application can directly judge the parallelizability and maximum parallelism of the computing region, and there will be no situation where different observation data interfere with each other and cannot be calculated in parallel during multi-core computing. In addition, the present application does not need to improve the ensemble Kalman filter algorithm, and is convenient to implement. Description of the Drawings
[0026] The following will further illustrate the present application in conjunction with the drawings. In the drawings:
[0027] Figure 1 is a schematic diagram of the influence region of the observation data;
[0028] Figure 2 is a flowchart of the acceleration method for the ensemble Kalman filter algorithm assimilation calculation according to an embodiment of the present application;
[0029] Figure 3 is a schematic diagram of dividing the computing region into multiple sub-regions according to an embodiment of the present application;
[0030] Figure 4 is a schematic diagram for parallelly performing assimilation calculations on the first set of non - adjacent sub - regions in the calculation area shown in Figure 3 ;
[0031] Figure 5 is Figure 4 a schematic diagram of the influence area of the observed data within the first set of non - adjacent sub - regions shown in
[0032] Figure 6 is Figure 5 a schematic diagram of the influence area of the observed data at the boundaries of the sub - regions shown in
[0033] Figure 7 is a schematic diagram for parallelly performing assimilation calculations on Figure 3 the second set of non - adjacent sub - regions in the calculation area shown in
[0034] Figure 8 is a schematic diagram for parallelly performing assimilation calculations on Figure 3 the third set of non - adjacent sub - regions in the calculation area shown in
[0035] Figure 9 is a schematic diagram for parallelly performing assimilation calculations on Figure 3 the fourth set of non - adjacent sub - regions in the calculation area shown in
[0036] Figure 10 is a schematic diagram of the logical structure of an acceleration device for assimilation calculations of the ensemble Kalman filter algorithm according to an embodiment of the present application;
[0037] Figure 11 is a schematic diagram of the logical structure of an electronic device according to an embodiment of the present application. Detailed implementation manners
[0038] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Moreover, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.
[0039] In order to more clearly illustrate the technical solution of the present application, the present application first introduces the process of the ensemble Kalman filter algorithm. Define N as the number of ensemble members, the background ensemble members are is the ensemble mean, X′ is the background field perturbation, is the background error covariance, H is the observation operator, y o is the observation vector, R is the observation error covariance, is the Kalman gain matrix, X aFor the analysis field, the ensemble Kalman filter algorithm is generally divided into the following four steps:
[0040] (1) Calculate the perturbation X′ of the background field
[0041]
[0042] (2) Obtain the background error covariance matrix
[0043]
[0044] (3) Calculate the Kalman gain
[0045]
[0046] (4) Obtain the analysis field X a
[0047]
[0048] The above steps (2)-(4) are the calculation processes for individual observation data. By continuously looping steps (2)-(4) until all observation data are calculated, the assimilation calculation process of the ensemble Kalman filter algorithm is completed. For the calculation process of the ensemble Kalman filter algorithm, further reference can be made to the following literature:
[0049] Zhuang Zhaorong. Design of Ensemble Kalman Filter Data Assimilation System and Its Application in Ensemble Forecasting [D]. Beijing: Graduate School of Chinese Academy of Sciences, 2007, 91:92.
[0050] During the implementation of the ensemble Kalman filter algorithm, to avoid difficult solutions, the observation data enters sequentially and the assimilation is gradually completed. During the sequential entry of the observation data, the observation data with overlapping influence ranges cannot enter the assimilation area for calculation simultaneously, while the observation data with non-overlapping influence ranges can be calculated in parallel. As Figure 1 shown, the black dots in the figure are the positions of certain observation data, and the gray areas represent the ranges where the observation data affect the calculation grid points. Among them, the influence areas of the two observation data shown in the left figure a) overlap, and at this time, these two observation data cannot be calculated simultaneously, while the influence areas of the two observation data in the right figure b) do not overlap and can be calculated simultaneously.
[0051] Based on this, to shorten the calculation time and improve the calculation efficiency, this application proposes an acceleration method for the assimilation calculation of the ensemble Kalman filter algorithm. Figure 2 The flowchart of the acceleration method 10 for the assimilation calculation of the ensemble Kalman filter algorithm according to an embodiment of this application is shown. As Figure 2 shown, this acceleration method 10 includes the following steps:
[0052] Step S11: Divide the calculation area into multiple sub-areas according to the influence radius of the observation data.
[0053] The influence radii of different assimilation data are different. Usually, a suitable influence radius can be determined through multiple tests, and then the entire calculation area is divided into multiple sub-areas according to this influence radius, so that the length from the center of each sub-area to its boundary is not less than the influence radius of the observation data, for example, it can be equal to or slightly larger than the influence radius. In the preferred embodiment, as shown in Figure 3 shown, the simplest division method is to divide the calculation area into multiple square sub-areas in a checkerboard pattern, and the side length of each square sub-area 40 is equal to the influence diameter of the observation data. Similarly, each square sub-area is also composed of multiple grids, and the number of grids in each row and each column is the smallest integer upper bound of the ratio of the influence diameter to the grid step size. Assume that the grid step size of the calculation area is h and the influence diameter is d, then the number of grids in a row inside the sub-area 40 is n = [d / h] + 1, where [] represents taking the integer. In different embodiments of the present application, the shape of the sub-area is not limited to the square shown in the figure, and can also be divided into other shapes such as circles.
[0054] Step S12: Select a group of non-adjacent sub-areas from the multiple sub-areas, perform the assimilation calculation of the group of non-adjacent sub-areas in parallel, and synchronously update the adjacent sub-areas of each non-adjacent sub-area, and repeat the execution until the assimilation calculation of all the multiple sub-areas is completed.
[0055] Taking Figure 3 the calculation area shown as an example, in the above step S12, it can start from the first sub-area 41 of the coordinate system of the calculation area, and select the first group of non-adjacent sub-areas, such as Figure 4 shown by the dark gray sub-areas in, and perform the assimilation calculation of the first group of non-adjacent sub-areas in parallel. Since the side length of each sub-area is equal to the influence diameter of the observation data, the observation data at the boundary of the Figure 4 dark gray sub-areas in can exactly affect half of the position of the white sub-areas and does not include the boundary (the influence at the boundary is 0), as shown by the dotted line in Figure 5 shown. In this way, the observation data inside the dark gray sub-areas can be assimilated simultaneously, so 9 processes with the same number as the first group of dark gray sub-areas can be used to perform the assimilation calculation of the first group of dark gray sub-areas in parallel. When performing the assimilation calculation, for the observation data inside each sub-area, sequential assimilation calculation or parallel assimilation calculation can be performed. When calculating to the boundary of the dark gray sub-areas, at this time, these observation data will synchronously affect the grids inside the white sub-areas. Figure 6It shows the position of the white sub-region affected by the observed data at the black boundary when the observed data is in the first dark gray sub-region. Therefore, in the assimilation calculation process, in addition to updating the dark gray sub-region, it is also necessary to synchronously update the values of the grid points in the white sub-region adjacent to each dark gray sub-region.
[0056] After the observed data of the first set of non-adjacent dark gray sub-regions is traversed, then starting from the second sub-region 42 adjacent to the right of the aforementioned first sub-region 41, a second set of non-adjacent sub-regions can be selected, such as Figure 7 shown by the hatched sub-region, and the assimilation calculation of the second set of non-adjacent sub-regions is executed in parallel. Similarly, when performing the assimilation calculation, for the observed data within each sub-region, sequential assimilation calculation or parallel assimilation calculation can be performed, and the adjacent sub-regions are synchronously updated.
[0057] After the observed data of the second set of non-adjacent sub-regions is traversed, then starting from the third sub-region 43 adjacent to the lower side of the aforementioned first sub-region 41, a third set of non-adjacent sub-regions can be selected, such as Figure 8 shown by the cross-hatched sub-region, and the assimilation calculation of the third set of non-adjacent sub-regions is executed in parallel. Similarly, when performing the assimilation calculation, for the observed data within each sub-region, sequential assimilation calculation or parallel assimilation calculation can be performed, and the adjacent sub-regions are synchronously updated.
[0058] After the observed data of the third set of non-adjacent sub-regions is traversed, then starting from the fourth sub-region 44, a fourth set of non-adjacent sub-regions can be selected, such as Figure 9 shown by the vertically hatched sub-region, and the assimilation calculation of the fourth set of non-adjacent sub-regions is executed in parallel. Similarly, when performing the assimilation calculation, for the observed data within each sub-region, sequential assimilation calculation or parallel assimilation calculation can be performed, and the adjacent sub-regions are synchronously updated.
[0059] Through the above process, the assimilation calculation of all sub-regions in the calculation area can be completed. Taking Figure 3 the shown calculation area as an example, according to the acceleration method of the above embodiment of the present application, 9 processes can be used for calculation simultaneously. If the observed data is evenly distributed, theoretically a 9-fold speedup can be achieved. In addition, it should be noted that the calculation order of the above four sets of sub-regions is arbitrary. The calculation can be performed in the order of the dark gray sub-region, the hatched sub-region, the cross-hatched sub-region, and the vertically hatched sub-region, or in the order of, for example, the cross-hatched sub-region, the vertically hatched sub-region, the dark gray sub-region, and the hatched sub-region. Although the multiple sub-regions within each set of sub-regions are calculated in parallel, the calculation of these four sets of sub-regions must be carried out step by step and cannot be carried out simultaneously, which is determined by the ensemble Kalman filter algorithm itself.
[0060] According to the acceleration method for the assimilation calculation of the ensemble Kalman filter algorithm in the above embodiments of the present application, the calculation region is divided into multiple checkerboard-shaped sub-regions according to the influence radius of the observation data. The observation data inside non-adjacent sub-regions can be calculated simultaneously to achieve parallelism, shorten the calculation time, and improve the calculation efficiency.
[0061] Based on the acceleration method for the assimilation calculation of the ensemble Kalman filter algorithm introduced above, the present application also proposes an acceleration device for the assimilation calculation of the ensemble Kalman filter algorithm. Figure 10 The logical structure diagram of an acceleration device 20 for the assimilation calculation of the ensemble Kalman filter algorithm according to an embodiment of the present application is shown. Refer to Figure 10 As shown, the acceleration device 20 includes a division module 21 and a parallel calculation module 22. Among them, the division module 21 is used to divide the calculation region into multiple sub-regions according to the influence radius of the observation data. The length from the center of each sub-region to its boundary is not less than the influence radius of the observation data. In a specific embodiment, the division module 21 divides the calculation region into multiple square sub-regions, and the side length of each square sub-region is equal to the influence diameter of the observation data. The parallel calculation module 22 is used to select a group of non-adjacent sub-regions from the multiple sub-regions, perform the assimilation calculation of the group of non-adjacent sub-regions in parallel, and synchronously update the adjacent sub-regions of each non-adjacent sub-region, and repeat the execution until the assimilation calculation of all the multiple sub-regions is completed. In a specific embodiment, the parallel calculation module 22 can first select a group of non-adjacent sub-regions from the first sub-region of the coordinate system of the calculation region to perform the assimilation calculation in parallel, and then select other groups of non-adjacent sub-regions from the adjacent sub-regions of the first sub-region to perform the assimilation calculation in parallel until the assimilation calculation of all the multiple sub-regions is completed. For the observation data inside each sub-region, sequential assimilation calculation or parallel assimilation calculation can be performed, and the adjacent sub-regions are synchronously updated. For the specific implementation of the division module 21 and the parallel calculation module 22, reference can be made to the specific descriptions of step S11 and step S12 of the acceleration method 10 for the assimilation calculation of the ensemble Kalman filter algorithm described above, and details will not be elaborated here.
[0062] Based on the acceleration method 10 for the assimilation calculation of the ensemble Kalman filter algorithm described above in the present application, the present application also proposes an electronic device 30. Refer to Figure 11 As shown, the electronic device 30 includes a processor 31 and a memory 32, and the processor 31 and the memory 32 are communicatively connected. The memory 32 stores a computer program, and when the computer program is executed by the processor 31, the processor 31 implements the acceleration method 10 for the assimilation calculation of the ensemble Kalman filter algorithm in the above embodiments of the present application.
[0063] The present application also provides a computer-readable storage medium storing a computer program, which when executed by a processor implements the acceleration method 10 for the ensemble Kalman filter assimilation calculation in the foregoing embodiments of the present application.
[0064] The foregoing are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included within the protection scope of the present application.
Claims
1. An acceleration method for the assimilation calculation of the ensemble Kalman filter algorithm, characterized in that, The method includes the following steps: S11. Divide the calculation area into a plurality of square sub-areas in a checkerboard pattern according to the influence radius of the observation data, and the side length of each square sub-area is equal to the influence diameter of the observation data; S12. Select a first group of non-adjacent sub-areas from the plurality of sub-areas, perform the assimilation calculation of the first group of non-adjacent sub-areas in parallel, and synchronously update the adjacent sub-areas of each non-adjacent sub-area in the first group of non-adjacent sub-areas. Then, after the observation data of the first group of non-adjacent sub-areas is traversed, select a second group of non-adjacent sub-areas from the adjacent sub-areas of the first group of non-adjacent sub-areas, perform the assimilation calculation of the second group of non-adjacent sub-areas in parallel, and repeat this process until the assimilation calculations of all the plurality of sub-areas are completed.
2. The acceleration method according to claim 1, wherein The step S12 includes: Starting from the first sub-area of the coordinate system of the calculation area, select a first group of non-adjacent sub-areas and perform the assimilation calculation in parallel; Starting from the adjacent sub-areas of the first sub-area, select other groups of non-adjacent sub-areas and perform the assimilation calculation in parallel until the assimilation calculations of all the plurality of sub-areas are completed.
3. The acceleration method according to claim 1, wherein In the step S12, sequential assimilation calculation or parallel assimilation calculation is performed on the observation data inside each sub-area, and the adjacent sub-areas are synchronously updated.
4. An acceleration device for assimilative calculation of the ensemble Kalman filter algorithm, characterized in that, It includes: A division module for dividing the calculation area into a plurality of square sub-areas in a checkerboard pattern according to the influence radius of the observation data; A parallel calculation module for selecting a first group of non-adjacent sub-areas from the plurality of sub-areas, performing the assimilation calculation of the first group of non-adjacent sub-areas in parallel, and synchronously updating the adjacent sub-areas of each non-adjacent sub-area in the first group of non-adjacent sub-areas. Then, after the observation data of the first group of non-adjacent sub-areas is traversed, select a second group of non-adjacent sub-areas from the adjacent sub-areas of the first group of non-adjacent sub-areas, perform the assimilation calculation of the second group of non-adjacent sub-areas in parallel, and repeat this process until the assimilation calculations of all the plurality of sub-areas are completed.
5. The acceleration device according to claim 4, characterized in that, The parallel calculation module is further used to select a first group of non-adjacent sub-areas from the first sub-area of the coordinate system of the calculation area and perform the assimilation calculation in parallel, and select other groups of non-adjacent sub-areas from the adjacent sub-areas of the first sub-area and perform the assimilation calculation in parallel until the assimilation calculations of all the plurality of sub-areas are completed.
6. The acceleration device according to claim 4, characterized in that, The parallel calculation module performs sequential assimilation calculation or parallel assimilation calculation on the observation data inside each sub-area and synchronously updates the adjacent sub-areas.
7. An electronic device, comprising a processor and a memory, the memory storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the acceleration method for the assimilation calculation of the ensemble Kalman filter algorithm as described in any one of claims 1-3.
8. A computer-readable storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a processor, it implements the acceleration method for the assimilation calculation of the ensemble Kalman filter algorithm as described in any one of claims 1-3.